05 - Raster manipulation

Welcome to Bulgaria

Throughout these exercises, you will be working with two different types of satellite imagery for the Kazanlak Valley in central Bulgaria. The images are provided by the JICA and GeoEye Foundation respectively:

Aster contains the elevation values for the area.
IKONOS image is a 4-band satellite image covering ca 150 sq km of the Kazanlak Valley with red, green, blue and infrared band, captured in 2001. As it is a fairly large orthophoto, I divided it into two halves with each half at 2 GB and placed them in ScienceData. You can download the full-resolution images manually from public www.sciencedata.dk folder, or directly with with file.download() using these direct links for West and East respectively. In exercise 6 you will play with a slightly reduced East image, which is provided as part of Github data folder.

You will learn to work with these rasters and plot vector data over them.

Task 1: Access raster data values

Raster data can be very big depending on the extent and resolution (grid size). In order to deal with this the raster() and brick() functions are designed to only read in the actual raster values as needed. To show that this is true, you can use the inMemory() function on an object and it will return FALSE if the values are not in memory. If you use the head() function, the raster package will read in only the values needed, not the full set of values. The raster values will be read in by default if you perform spatial analysis operations that require it or you can read in the values from a raster manually with the function getValues().

Instructions

Activate raster and rgdal library
Use GDALinfo() to inspect the properties of the Aster image raster in the data folder (the file name is “Aster.tif”). What can you learn from this inspection about its bands, resolution, and projection?
Read in the Aster image. Review Week 02 raster loading if unsure about how.
Use the inMemory() function on the elevation object to determine if the data has been read in.
Use the head() function to look at the first few values from the elevation raster.
Use the getValues() function on the elevation object to read in all the data.
Use the hist() function to create a quick histogram of the elevation values. Note the pile of values near -9999, these should be NA (any idea why?) and we will address this later.

# Library
pacman::p_load(raster, rgdal, tidyverse)

GDALinfo("../data/Aster.tif")

rows        2484 
columns     2592 
bands       1 
lower left origin.x        313183.7 
lower left origin.y        4676606 
res.x       30 
res.y       30 
ysign       -1 
oblique.x   0 
oblique.y   0 
driver      GTiff 
projection  +proj=utm +zone=35 +datum=WGS84 +units=m +no_defs 
file        ../data/Aster.tif 
apparent band summary:
  GDType hasNoDataValue NoDataValue blockSize1 blockSize2
1  Int16          FALSE           0          1       2592
apparent band statistics:
   Bmin Bmax     Bmean      Bsd
1 -9999 2347 -3238.349 5106.829
Metadata:
AREA_OR_POINT=Area 
TIFFTAG_DATETIME=2009:04:10 19:05:46 
TIFFTAG_DOCUMENTNAME=created by Program:silcast_cmm 1.09/Ortho at 
TIFFTAG_IMAGEDESCRIPTION=SILC TIFF 
TIFFTAG_RESOLUTIONUNIT=2 (pixels/inch) 
TIFFTAG_SOFTWARE=IDL 6.3, Research Systems, Inc. 
TIFFTAG_XRESOLUTION=100 
TIFFTAG_YRESOLUTION=100

# Read in the elevation layer
elevation <- raster("../data/Aster.tif")

# Check if the data is in memory
inMemory(elevation)

[1] FALSE

# Use head() to peak at the first few records
head(elevation)

       1     2     3     4     5     6     7     8     9    10    11    12
1  -9999 -9999 -9999 -9999 -9999 -9999 -9999 -9999 -9999 -9999 -9999 -9999
2  -9999 -9999 -9999 -9999 -9999 -9999 -9999 -9999 -9999 -9999 -9999 -9999
3  -9999 -9999 -9999 -9999 -9999 -9999 -9999 -9999 -9999 -9999 -9999 -9999
      13    14    15    16    17    18    19    20
1  -9999 -9999 -9999 -9999 -9999 -9999 -9999 -9999
2  -9999 -9999 -9999 -9999 -9999 -9999 -9999 -9999
3  -9999 -9999 -9999 -9999 -9999 -9999 -9999 -9999
 [ reached getOption("max.print") -- omitted 7 rows ]

# Use getValues() to read the values into a vector
vals <- getValues(elevation)

# Use hist() to create a histogram of the values
hist(vals)

Congratulations! You now know that the raster package only reads in raster values as needed to save space in memory. You can get the values manually using the getValues() function.

Now, a new question arises, why are so many values encoded as -9999? Are we in the Mariana Trench all of sudden?

Task 2: Change values: handle missing or bad data values in rasters

There are many situations where you might need to recode raster values. You may want to change the outlier values to NA for example. In the raster package, reclassification is performed with the reclassify() function.

In the elevation raster you’ve worked with the values are meters above sea level and are supposed to range between 0 and 2500. Anything below 0 should be an NA. In this exercise you will assign any values below 0 to NA.

Instructions

Check that the package raster and the object elevation are loaded in your workspace.
Plot the elevation raster using plot().
Set up a three-column matrix with the cbind() function and values -10000, 0, NA.
Use the matrix and reclassify() to reclassify values below 0 to NA. You will need to use the argument rcl.
Plot the reclassified elevation layer to confirm there are no values below 0.

# Plot the elevation layer to see the legend values
plot(elevation)

dim(elevation)

[1] 2484 2592    1

# Set up the matrix
vals <- cbind(-10000, 0, NA) 

# Reclassify 
elevation_reclass <- reclassify(elevation, rcl = vals)

# Plot again and confirm that the legend range is 0 - 2400
plot(elevation_reclass)

Good work! Knowing how to reclassify rasters will come in handy. When you get a chance you should review the help for reclassify() particularly the part that discusses how to specify the rcl argument. The three-column approach from this exercise is most common but there are other approaches.

Task 3: Limit rasters to focus areas

Mask and crop are similar operations that allow you to limit your raster to a specific area of interest. With mask() you essentially place your area of interest on top of the raster and any raster cells outside of the boundary are assigned NA values. With crop() you limit the extent of your raster to that of your focus area.

The Aster image covers a large area, but we are primarily interested in the areas surveyed by archaeologists, the first of which is the cluster of burial mound points and second, large survey polygons registered by local archaeologists during field survey. In this exercise you will use the vectors of mounds and survey units to crop and mask the elevation raster.

Note: In the past the raster package did not support sf objects. This should not happen now, but if you encounter difficulty with raster:vector interactions, it helps to know that you can convert the vector to Spatial object with, for example, as(input, "Spatial").

Instructions I

Load sf library
Create mounds object from the shapefile “KAZ_mounds.shp”. For a refresher on vector loading, check out Week 02 instructions.
Verify that mounds CRS matches the elevation and fix with st_transform() if not.
Create a bounding box around the mounds with st_make_grid() following Week 03 guidelines.
Plot the mounds_bb and the mounds to see how these two objects relate.
Crop the elevation layer by the mounds object and create a smaller elev object
Plot the elev and mounds and the mounds_bb together.

# YOUR CODE
pacman::p_load(sf)

mounds <- read_sf("../data/KAZ_mounds.shp")
st_crs(mounds) == crs(elevation)

[1] FALSE

mounds <- st_transform(mounds, crs(elevation))

mounds_bb <- st_make_grid(mounds)

elev <- raster::crop(elevation_reclass, mounds)

plot(elev); plot(mounds$geometry, add=T); plot(mounds_bb, add=TRUE)

Instructions II

Create survey object from shapefile called “KAZ_units.shp”
Project the survey object to match the new elev raster with st_transform()if needed.
Compute the area of the survey with st_area() and save this object as areas. What units are these?
Filter the survey units to only those above 30000 square meters with the filter() function. You will need to wrap areas in unclass(). Save as survey_big. Remember to have the tidyverse or dplyr library attached for filter() to work properly. Also sf might interfere so specify the dplyr::filter() if needed.

# Read in the survey object
survey <- read_sf("../data/KAZ_units.shp") %>% 
  st_transform(crs(elevation_reclass))

# Compute the area of the survey
areas <- st_area(survey)

# Filter to survey with areas > 30000
survey_big <- dplyr::filter(survey, as.vector(areas) > 30000)

Instructions III

Review the plot of elev raster.
Plot the geometry of the survey_big over it.
Mask the elev layer with survey_big and save as elevation_mask. This may take a couple of seconds.
Review the plot of elevation_mask.

# Plot the elevation raster
plot(elev)

# Plot the geometry of survey_big
plot(elev); plot(st_geometry(survey_big), add=T)

# Mask the elev layer with survey_big and save as elevation_mask
elevation_mask <- mask(elev, mask = survey_big)

# Plot elevation_mask -- this is a raster!
plot(elevation_mask)

Nice! You ensured that layers had the same CRS, you cropped raster, computed bounding boxes and areas for masking and filtered the data. Finally, you used mask() to mask the elevation raster to the large survey units.

Question:

1. What extent does the elevation raster default to after cropping by mounds vs by masking by the large polygons of survey_big? They have approximately the same extend - however, it simply overlays in the first case, whereas the second case only shows the elevation of the mounds.

2. Why do we not use the mounds bounding box to crop the elevation raster? It seems like they have the wrong datatype - it at least throws an error when you try.

Task 4: Reduce the raster grid cell size using aggregate

Rasters, such as ortophotos, terrain models, or aggregated rasters for large area, often come is resolutions far greater than you need. Up till now you have played with fairly small, processed rasters. Now you are getting a taste of the real thing. To reduce the computational load when running analyses, or when developing the right approach, you might wish to use a reduced resolution raster.

The function to reduce resolution in rasters is aggregate() which, as you might guess, aggregates grid cells into larger grid cells using a user-defined function (for example, mean or max). The function used to aggregate the values is determined by the fun argument (the default being mean) and the amount of aggregation is driven by the fact argument (the default being 2).

Instructions

Load the East half of the two IKONOS images of the Kazanlak Valley from (“KazE.tif”).
Plot the file you read in with the appropriate function for a multi-band raster.
Determine the raster resolution using res() and number of raster cells in the layer with ncell().
Aggregate the IKONOS image using the default for fun and a factor of 10 and save the new raster to east_small.
Plot the new raster with plot() for comparison to the old version.
Determine the new raster resolution and the number of raster cells.

# Load the IKONOS raster
east <- brick("../data/KazE.tif") 

# Plot the IKONOS raster
plotRGB(east, stretch="lin", r=4)

# Determine the raster resolution
raster::res(east)

[1] 10 10

# Determine the number of cells
ncell(east)

[1] 2374596

# Aggregate the raster
east_small <- aggregate(east, fact = 10)

# Plot the new east layer
plotRGB(east_small, stretch="lin", r=4)

# Determine the new raster resolution
raster::res(east_small)

[1] 100 100

# Determine the number of cells in the new raster
ncell(east_small)

[1] 23940

Lovely job! In this example you read in a raster and then converted it to a lower resolution raster to save on the size of the object and ultimately computation power required. In this example, the raster was not too big to begin with so perhaps aggregating would not be necessary but for big rasters such as you will see next week, this can be a big help and a necessity.

Task 5: Extract raster values by location

Beyond simply masking and cropping you may want to know the actual cell values at locations of interest. You might, for example, want to know the elevation at each mound location or perhaps the mean elevation within the large survey units. This is where the extract() function comes in handy.

Usefully, and you’ll see this in a later analysis, you can feed extract() a function that will get applied to extracted cells. For example, you can use extract() to extract raster values by point, line, polygon, or neighborhood and with the fun = mean argument it will return an average cell value by neighborhood.

Instructions

Ensure mounds are still in memory.
Use the raster function extract() to determine the elevation at each of the mounds. Assign the extracted value into a elev column in the mounds object. Beware that the extract() function exists across multiple packages, so it’s wise to use raster:: in front of it.
Look at the mounds elevation column through the plot() function as well as histogram. Do the extract() results make sense? The elevation layer values represent meters above sea level.

# Extract the elevation values at the sites
mounds$elev  <- raster::extract(elevation, mounds)

# Look at the sites and extraction results
plot(mounds["elev"])

hist(mounds$elev)

# Makes sense that nothing is ridicolously high or below sea level (for now)

Great! raster::extract() is a very useful tool. It can be used with polygons as well as points, and the result can be written out as a separate vector or added to an existing object as a column.

Task 6: Raster math with overlay

You will now use the elevation layer and an “prominence” layer. Prominence measures what percentage of surrounding cells are visible from any given location in the raster. A high percentage value thus indicates a prominent point, while a low percentage indicates a low-lying location with poor inter-visibility. Archaeologists assert that mounds had a good field of visibility, so let’s see whether they are mostly right :)

What you will do in this exercise is essentially identify the most prominent locations among the registered archaeological mounds by finding areas that have both a high percentage of prominence and high elevation. You will use two rasters and you will define an overlay function f to do the raster math with. Note that you can only do raster math on two rasters of the same extent, so make sure you align the elevation and prominence accordingly with crop() function

Instructions

Make sure elevation object exists in memory (“Aster.tif”, it is a single-band raster).
Read in the prominence raster layer from “prominence.tif”; it is also a single-band raster.
Plot the prominence object. Do the legend units make sense?
Specify function f, where you look for elevation values over 400 msl and below 650m (mounds don’t appear above that elevation) and prominence values over 60%
Call overlay() on elevation and prominence. Set the fun argument to f.
If you have not cropped the two rasters to the same extent yet, you can do so inside the overlay() function

# Check in on the elevation and read in prominence layer
prominence <- raster("../data/prominence1500m.tif") %>% 
  raster::projectRaster(crs=crs(elevation))

# Plot prominence
plot(prominence)

# Function f with 2 arguments and the raster math code
f <- function(r1, r2) {
    (400 < r1) & (r1 < 650) & (r2 > 60)
}

# Align the extent of the two rasters with crop()
c_prominence <- crop(prominence, elev)

# Do the overlay using f as fun. 
elevation_prom_overlay <- overlay(elev, c_prominence, fun = f)

# Plot the result (low elevation and high prominence areas)
plot(elevation_prom_overlay)

Congratulations! You’ve now learned to perform raster math using the raster function overlay(). You limited to areas with > 400m & <650m elevation and > 60% prominence, these areas should harbor the most prominent mounds (or defense-ready areas if associated with settlements) in Kazanlak.

Questions:

3. What is the actual value range in the prom_el_overlay raster? it seems to be [0, 1] 4. What area is covered by each cell?

raster::res(elevation_prom_overlay)

The resolution is 30m^2 # Task 7 - HOMEWORK: Inspect the most prominent mounds

In the above exercise, we produced an elevation-prominence overlay. Mounds and other sites that sit in this overlay enjoy a strategic position vis-a-vis the rest of the valley. Which ones are they, however, and what are their real prominence values? It is hard to see at the scale of the valley and it would be good to pull the sites out.

Find the mounds that enjoy the most prominent locations as well as those that feature in the elevation-prominence overlay raster. Produce a list of the ID numbers (TRAP_Code) of all the overlay mounds and 25 most prominent mounds and plot them (expressing their prominence somehow) .

Instructions

Find or recreate the mounds sf object, prominence and prom-el-overlay rasters.
Plot the prom-el-overlay and the mounds on top of each other to check visual overlap. Do the same with prominence raster and mounds.
Extract values from the elevation-prominence overlay raster and from the prominence raster at mounds and write them to two columns:

– call the first column prom_el_overlay – name the second column prominence
Make an object of mounds that sit within these strategic high-visibility locations. How many are there?
Make an object of 25 mounds with the highest prominence values. Which TRAP_Code ids are included?
Plot both these sets of mounds in mapview() and compare their locations.

# Plot prominence/elevation and mounds
plot(elevation_prom_overlay); plot(mounds, add=T)

# Plot prominence and mounds
plot(prominence); plot(mounds, add=T)

# Extracting
mounds$prom_el_overlay <- raster::extract(elevation_prom_overlay, mounds)
mounds$prominence <- raster::extract(prominence, mounds)

strategic_mounds <- mounds %>% 
  dplyr::filter(prom_el_overlay == 1)

prominent_mounds <- mounds %>% 
  slice_max(prominence, n=25)

print(prominent_mounds$TRAP_Code)

 [1] 2002 3112 3105 4057 4058 3722 4059 3417 4056 3276 3024 1020 4087 1019 3026
[16] 2028 4066 5002 4086 1071 3275 2003 2004 3703 3420

mapview::mapview(strategic_mounds) + 
  mapview::mapview(prominent_mounds, zcol="prominence")

Question:

5. How do the mounds with high prom_el_overlay values differ from those with high prominence? Most of the high prominence mounds seem to also be highly elevated, which makes a lot of sense # Task 8 - OPTIONAL HOMEWORK: Sum raster values across polygons - assessing prominence of past settlements

Previously you have extracted values from computed rasters at points. More often, you will need to extract values using more complex vector shapes where you will need to decide how to summarize the raster values that fall in. In this exercise, you will load the layer of ancient settlements for the Kazanlak Valley and sum the overlapping elevation_prom_overlay values within the settlement polygons. Then you will find out which of the sites contained at least 0.5 ha of strategic high-visibility area.

Instructions

Create a sites sf object by reading in polygons from “KAZ_nuclei.shp”
Plot the elevation-prominence overlay and the sites on top of each other
Use mask function to create a sites_mask raster, where raster cells under sites’ polygons retain their original values, while cells outside the polygons are turned to NA.
Use crop function to crop sites_mask raster to the extent of the sites’ polygons.
Plot the result ( I recommend mapview())
Produce a list of sites that have at least 0.5 ha (ie. 6 grid cells) within these strategic high-visibility locations. (Hint:you can first extract a list of overlapping sites and then use sapply() to filter out the ones > 6)

# Read in the site polygons
sites <- ______("data/KAZ_nuclei.shp")

# Plot the sites
plot(elevation_prom_overlay);plot(__________)


# Create a raster mask from the elevation_prom_overlay using the sites
sites_mask <- _____________

# Crop the raster mask to the sites
sites_strat <- _____________

# Plot and compare the results
plot(sites_mask)
plot(sites_strat)


# Produce a list of the most prominent sites

Lovely work! Manipulating rasters and mastering raster-vector interactions is the bread and butter of spatial data wrangling :)